Music research papers
random trip report

Journals and conferences

Sadly, the following use the (horrible, rip-off) "pay to view" model:


I subdivide this as follows:

"Nuance analysis": studying a corpus of performances.

  • Statistical: study the statistics of the nuance.
  • Model-based: find specific parameterized functions that fit the data.

"Nuance generation": adding nuance to a rendition.

  • AI-based: train a neural net with a corpus of performances.
  • Algorithmic: add nuance with deterministic rules based on pitch contour etc.
  • Explicit: a human specifies nuance.

These are related; e.g. a system for explicit nuance generation provides a set of primitives, and ideally these would be chosen based on model-based analysis.

Manfred Clynes and Superconductor

Manfred Clynes did work on explicit (and possibly algorithmic) nuance generation, in the context of a Windows program called SuperConductor (not related to SuperCollider AFAIK). Superconductor does audio synthesis (e.g. of bowed string sounds). It provides control (through a GUI) of high-level (e.g. crescendi) and note-level (e.g. vibrato) nuance. It supports nonlinear curves (exponentials, cubic splines?). It looks rigid and limited. It comes with a few pieces by "The Great Composers" (in a proprietary format) that you can add nuance to. It can also import/export MIDI.

There are a bunch of examples on YouTube, mostly from string quartets. These are not bad - they have nuance with appropriate general properties. However, the nuance is excessive, and it's clear within a few seconds that it's computer-generated.

This is based on Clynes' theory of acoustic aesthetics/emotion called "Sentic forms", which holds that composers have characteristic "hierarchical pulses". This is pretty vague and speculative. Also, all of Clynes' stuff (including his self-authored Wikipedia page) has an uncomfortably high level of self-promotion.

Papers by Clynes:

Bruno Repp


  • MAESTRO: a corpus of piano performances (MIDI file and audio).


Papers that study performance nuance, and show actual data.

Taylor & Francis: no Semantic scholar: sometimes researchgate: sometimes yes (arabic)

Copyright 2023 © David P. Anderson